Tim,

I agree with you. Your critics on MI is just about what inspired us to
do the Traits work. Have a look at the latest draft of the Traits paper
(http://www.iam.unibe.ch/~schaerli/traits_draft/traitsDraft.pdf) and you
will find a detailed description of good reasons why MI ddidn't make it
into recent languages. I hope that Traits are going to be more
successful ;)

Nathanael

> -----Original Message-----
> From: squeak-dev-admin at lists.squeakfoundation.org 
> [mailto:squeak-dev-admin at lists.squeakfoundation.org] On 
> Behalf Of Tim Rowledge
> Sent: Monday, November 18, 2002 7:38 PM
> To: squeak-dev at lists.squeakfoundation.org
> Subject: Re: Classes as Packages (was: Harvesting infrastructure)
> 
> 
> 
> > > And anyone that starts going on about multiple inheritance is 
> > > looking for a visit from one of the four horsement of the 
> > > apologists. :-)
> > 
> > I think this is a myth of multiple inheritance.  I think it can be 
> > tamed and used effectively.  See above.
> Lots of people have tried to convince of that in the last 
> twenty years and so far not a one has made a good enough case 
> to change my mind. It is certainly possible you may be the 
> first but don't hold your breath too long!
> 
> One of the classic old claims for MI was along the lines of - 
> well, I have implemented Toy and Truck and now I want a toy 
> truck so I should be able to inherit from both. Yippee, it's 
> obvious! BZZZZZZ.
> 
> Firstly, like many occasions where people make claims 
> congurent to this, it's a bad example. Truck, if implemented 
> in any serious manner, will have a great deal of state and 
> behaviour that is nothing whatosever to do with toy trucks. 
> Do toy trucks need to know about, say, service schedules, 
> fuel consumption, tyre pressure et al.? I doubt it. It's a bad mix.
> 
> Secondly, I've never yet been convinced that any proposed 
> implementation of MI can handle the tangle of possible 
> inheritance upstream. Where and how should one decide which 
> of the possible upward paths to follow? If following one path 
> leads to a DNU should you try another? And another? I think 
> it is easy to see a pretty nasty O(something horrible) 
> problem here. How would one specify which path to use it if 
> it supposed to be decided by the programmer? The old ST-80 MI 
> experiment used munged selectors, something like 'self 
> truck.size' which looked 'orrible. And probably allowed 
> something like 'self number.wibble' whcih wuldbe nonsensical.
> 
> Thirdly, it always seems more plausible to me to use 
> composition/delgation. ToyTruck would be a little better if 
> implemented as having a toy-ness ivar and a truck-ness ivar 
> and then _using_ them rather than inheriting them. I imagine 
> that this maps onto Traits reasonably well from my current 
> understanding. I would concur however that making it possible 
> to subclass traits would be (probably) useful; but let's not 
> go into whether traits should have trairts, at least just yet.
> 
> tim
> -- 
> Tim Rowledge, tim at sumeru.stanford.edu, 
> http://sumeru.stanford.edu/tim Useful > random insult:- He 
> donated his brain to science but they made an early withdrawal.
> 
>